[8/8] kudu git commit: cache: fix behavior on single-CPU systems

[da...@apache.org]

cache: fix behavior on single-CPU systems

On a system with only a single CPU, shard_bits_ would be set to 0. This
would then result in calculating 'hash >> (32 - 0)' which is undefined
behavior. With optimizations, this would turn into a no-op, and we'd end
up using the whole hash as the shard index, instead of 0, causing a
crash.

The fix is to widen the hash to uint64_t before shifting.

Tested by manually making NumCPUs return 1 and running cache-test.

Change-Id: I7809e5697df657a589b2ceae5c6d4edbf161b52a
Reviewed-on: http://gerrit.cloudera.org:8080/4535
Tested-by: Kudu Jenkins
Reviewed-by: Adar Dembo <ad...@cloudera.com>
(cherry picked from commit 69e65780843040c085ca34901803b97b59595c0e)
Reviewed-on: http://gerrit.cloudera.org:8080/4608
Reviewed-by: Todd Lipcon <to...@apache.org>


Project: http://git-wip-us.apache.org/repos/asf/kudu/repo
Commit: http://git-wip-us.apache.org/repos/asf/kudu/commit/40637ef8
Tree: http://git-wip-us.apache.org/repos/asf/kudu/tree/40637ef8
Diff: http://git-wip-us.apache.org/repos/asf/kudu/diff/40637ef8

Branch: refs/heads/branch-1.0.x
Commit: 40637ef8fc773dc47882334fcee760bbffcd09b4
Parents: 77769ad
Author: Todd Lipcon <to...@apache.org>
Authored: Sat Sep 24 19:35:43 2016 -0700
Committer: Dan Burkert <da...@cloudera.com>
Committed: Mon Oct 3 21:47:53 2016 +0000

----------------------------------------------------------------------
 src/kudu/util/cache.cc | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
----------------------------------------------------------------------


http://git-wip-us.apache.org/repos/asf/kudu/blob/40637ef8/src/kudu/util/cache.cc
----------------------------------------------------------------------
diff --git a/src/kudu/util/cache.cc b/src/kudu/util/cache.cc
index 964d93c..1ef16a0 100644
--- a/src/kudu/util/cache.cc
+++ b/src/kudu/util/cache.cc
@@ -395,7 +395,9 @@ class ShardedLRUCache : public Cache {
   }
 
   uint32_t Shard(uint32_t hash) {
-    return hash >> (32 - shard_bits_);
+    // Widen to uint64 before shifting, or else on a single CPU,
+    // we would try to shift a uint32_t by 32 bits, which is undefined.
+    return static_cast<uint64_t>(hash) >> (32 - shard_bits_);
   }
 
  public: